Detecting disease clusters: the importance of statistical power

Area of Science:

• Epidemiology
• Biostatistics
• Spatial Analysis

Background:

• Numerous statistical methods exist for analyzing disease excesses.
• Few studies compare these methods' effectiveness in detecting disease clusters.
• Statistical power is crucial for evaluating cluster detection methods.

Purpose of the Study:

• To compare the statistical power of two common space-time disease cluster detection methods.
• To evaluate methods based on sensitivity, false positive rates, and cluster structure assumptions.
• To assess performance with small case counts typical of public health surveillance.

Main Methods:

• Reviewed two disease cluster detection methods: cell occupancy models and interevent distance comparisons.
• Conducted power studies to assess the probability of rejecting the null hypothesis when false.
• Analyzed sensitivity (true positives), specificity (true negatives), and false positives.

Main Results:

• Both evaluated methods demonstrated low statistical power for small numbers of disease cases.
• This low power is particularly relevant for citizen reports to health departments.
• Methods are optimized for different null hypotheses, impacting their detection capabilities.

Conclusions:

• Commonly used disease cluster detection methods have limitations, especially with sparse data.
• Further research is needed to improve the power of statistical methods for public health surveillance.
• Careful consideration of the underlying hypothesis is necessary when selecting a cluster detection method.